Google's Suncatcher Satellite Sends AI-Written Haiku From Orbit

What was launched

The satellite went up on 1 October with Planet on the Transporter-18 rideshare mission with SpaceX. It flies at 800 km on a pole-to-pole route that maximises sun exposure. Each satellite carries dozens of Trillium Tensor Processing Units (TPUs). Google plans to link many satellites into AI compute constellations that would eventually act as large data centres. It will test that setup next year with a second satellite.

Engineers stress-tested the compute modules with Planet before launch. Launch subjects the hardware to up to 100 times the force of gravity. The team is now tracking how the chips cope with radiation, which can cause bit flips and silent data corruption. Google and Planet engineers reproduced these effects in lab testing.

The heat problem

Cooling is a major challenge. Space is cold, but without air heat does not simply dissipate. Suncatcher uses heat pipes, closed pumped fluid loops and flat panel radiators that emit heat as infrared radiation. TPUs can generate over 100,000 watts of heat per square metre, while the radiators dissipate only around 300 W/m2. As a result, the satellite has to pause every 15 minutes to shed excess heat.

A crowded race

Starcloud was the first to put AI in space last year, with a conventional NVIDIA H100 GPU that ran Gemma. Google argues its specialised TPUs will scale more efficiently. Solar panels in orbit can draw five to eight times as much power as ground-based ones. Elon Musk has also praised the idea. Other projects include Blue Origin's Project Sunrise, SpaceX's Starmind and the European Space Agency's ASCEND. Some people have raised environmental concerns.

"Everything seems too hard until someone tries it," said Planet's Eric Stevens, calling the project an experimental moonshot.